← All transcripts

Doing The Right Thing: Accountability In The Age Of AI Coding Summary Transcript, AI Summary & Key Points

InfoQ · 4 days ago · Science & Technology · 23:19 · EN

Watch on YouTube

Answer

Humans are accountable when AI writes the code: AI is a powerful tool, and the person wielding it is in charge and must answer for what goes wrong — blaming the agent is an abdication.

AI Summary

Hugo Venturini argues humans stay accountable for AI-generated code: AI is a powerful tool, but the person conducting it answers for what goes wrong. Pull requests are growing — stories of 2,000-line diffs and senior engineers with backlogs of 300 unreviewed requests — and the fix is not reviewing everything: automate verification (does it compile, pass tests, behave as intended), let trusted seniors merge unreviewed diffs, and reserve review for the heart of the product. Venturini draws a parallel to Microsoft's 2014 merger of the SWE and tester branches, which raised quality by making engineers accountable for the final product; today's role collapse around AI has the same shape but no quality gain yet, because people still blame the AI instead of feeling accountable. Leaders close that gap with transparency and psychological safety: his weekly Q&A at a broken-culture company — answering 'I don't know', admitting mistakes, taking anonymous questions via another manager — rebuilt trust over patient months. Accountability without blame lets people dare; culture change works at any level, including single teams, and ICs can hold similar influence through technical strength, humility and listening. Managers are amplifiers who belong to two teams and must think beyond their own.

Key Points

  • Doing the right thing means solving the user's problem, with each step — deciding, writing, testing, releasing — done right in itself; for code generation it means generating the proper code.
  • Humans are in charge and accountable when AI fails; AI is just another, very powerful tool, and doing the right thing for it means doing what the human asked.
  • Pull requests are growing — stories of 2,000-line diffs and a senior engineer with a 300-request backlog — a problem that predates AI; the 'I rewrote an entire product in one diff' flex is just for show.
  • Not everything needs review: verify what can be automated (compiles, passes tests, behaves as intended), let trusted seniors merge unreviewed diffs and own the fix, and review only the heart of the product for quality and knowledge sharing.
  • Microsoft's 2014 merger of software engineers and testers raised quality because engineers became accountable for the final product; today's AI-driven role collapse should do the same, but quality hasn't risen because people still blame the AI.
  • AI empowers people to do jobs they've never done — PM, architect, tester — but it answers the questions you ask, and being senior is asking the right questions; people need training in that skill.
  • People dare when they know they will be accountable but not blamed; empowerment without psychological safety produces nothing.
  • Case study — fixing a broken culture as CTO: weekly Q&As, leading by example with 'I don't know' and 'I made a mistake', naming solutions not people, anonymous questions routed through another manager so their origin can't be guessed; within months people asked hard questions and the organization improved.

Links mentioned

🔒 Full analysis locked

Unlock more videos and the full analysis

A credit unlocks one video's full analysis for good — the build steps, the tools and how each was used, the methods behind every use case. Pro opens the whole library instead, and raises how many videos you can analyse a day.

Unlock full analysis — free

Transcript

Searchable transcript of Doing The Right Thing: Accountability In The Age Of AI Coding Summary — InfoQ (23:19). Search for a phrase, then click its timestamp to jump straight to that moment in the video.

Captions sourced from the original video on YouTube, published by InfoQ. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.

00:00 How much authority should an AI agent have? At CUCONAI New York, learn how senior practitioners set permissions and evaluate agents in production. Use their experience to decide where your own agents need boundaries and checks before release. Join us December 15 and 16, 2026 at the Western Jersey City, Newport. Explore the program at new york.qcon.ai.

00:33 Good day folks. This is Shane Hasty for the InfroQue Engineering Culture Podcast. Today I'm sitting down with Hugo Venturini. Hugo, welcome. Thank you for taking the time to talk to us. >> Thank you, Shane, for having me. >> Now we're sitting in opposite sides of the world. I'm in New Zealand and you're somewhere in Europe. >> Yes, right now I'm in Spain in Valencia.

00:56 >> Nice. You've got a beautiful warm summer. I'm sitting in a pretty cold winter, but we'll get past that. My normal starting point with these conversations is who's humor? So, I'm a software engineer. I defended my PhD on the debugging of optimized code in 2008 and after that I joined the industry as a software engineer and I climbed the ladder you know one step after the other senior then eventually manager.

01:22 I was manager at Meta, Facebook at the time. I was accountable for the type checker and the compiler of the hack language. Eventually I decided to try to create my own company grew it then joined another organization was the CTO of quot the search engine 90 people in the organization from network engineers to AI researchers and right now I work in a software company startup company that deals with tooling for AI to help AI do the right things so to speak >> the right thing.

01:58 What do we mean when we say the right thing today? >> Somebody's making a decision about what should be provided as a product to users. Somebody's about to write the code. Somebody will test it. Somebody will make the decision to release this to prod and somebody will use it. It can be all different people or it can be the same person at every step.

02:22 Doing the right thing to me is solving the problem to a user. And then at every step you got doing the right thing of that step. And for a code generation it is generating the proper code to some extent. >> Now somebody made a decision but all too often today we're seeing the abdication of that but the agent did this. The co-generator did this. I think humans are in charge and humans should be accountable and are most of the time at least what I can see and that's the right thing.

03:00 I'm not minimizing what it does but AI is just another tool. It's a very powerful tool and don't get me wrong I still got these wow moments several times not a day but at first a week. And yet if it fails the one conducting this great tool is accountable is in charge and should answer what goes wrong. doing the proper thing is doing what the human asked it to do.

03:24 And when you want to generate code, code code has to be, you know, to the column, to the semicolon, to the character. And if it does not, many times the proper tooling can actually tell the AI how to loop and give the right feedback. So when it comes back to the human, a great deal of what could be automated has been automated. One of the things that we're seeing, and various studies are pointing to this, the size of pull requests are getting bigger and bigger.

03:52 I've heard stories of the 2,000 lines. A senior engineer who's got a backlog of over 300 pull requests, each of which is hundreds of lines long, waiting to be reviewed. First the size of pull request is a problem which does not rise with AI. It was already existing before and the first thing is asking yourself should I review everything sometime you got all that boiler plate we could already before AI automate the verification let me give it as if it compiles that's already a good thing if it passes all the tests

04:30 you've created if it behaves the way you wanted it do you really need to review the code that was already not the case before second thing is already before as well and I lived in organization that was working like this you don't have to go for a signature of a diff for every single div you do eventually people are good enough senior enough or do trust each other that says I can totally merge this diff without having anybody reviewed it and if it goes wrong I'm accountable for fixing it that'd be okay but I'm totally

05:04 sure and people trust me to merge it already and sometimes you got the heart of the product which you want to create and you need some people to review for two reasons First for quality, second for sharing the knowledge of course and sometimes it's really just boiler plate that you can totally merge it with AI big reviews always and you can still I believe not go for the who's got the biggest review but who's handy enough to get your AI to create the proper diff and the proper stack of diff.

05:32 So you got only want to review what the heart of the product is. I think with AI because of the wow effect some people are playing this game of the wow effect with look I have rewritten this entire big product in only one diff which is just for the show let's face it nobody does this and before at the group of engineers if anybody were doing this we were all telling him or her please help us review your code that's one thing and after that you got the number of divs and that's another problem how do you review from

06:05 number of If I don't have any good answer and I'm really eager to have anybody telling me how do you do this? So right now what we've decided in my team is we don't review everything. We were not reviewing everything before and we trust people with our own good judgments that this needs not being reviewed and we only ask proactively this I want you to review it.

06:24 I give you two of reasons for that and this I can merge it already. And so this minimize number of tips to be reviewed by other people but that's an open conversation. And I'm sure we all want to have like how do others do about this and please tell me. >> Well, we'll leave that as an open question for the audience to respond. So, what are the implications of this in terms of the broader elements of quality in our product and we think of the architectural frameworks and so forth.

06:55 How do we make sure that we are building the right stuff? I got a story about rising quality in an unexpected way, at least for some. I worked at Microsoft in 2014 when they decided to merge the software engineer branch with the testers branch. At the time, testers were doing two things. Some were not coding, some were coding. and the software engineers were writing software and handing it to the testers for testing and was going back and forth.

07:29 By removing the official testing step by some somebody else, it actually rised the quality of the software and the reason I believe is because it made software engineers accountable for the final product, not one step of the chain. Today I got the feeling that we are removing the software and everybody's becoming a tester via your AI because somebody you command some software which is written and then people become software but we are all again merging those two verticals as one and the difference is we don't see an

08:07 increase of quality yet and I believe it's because people are still blaming the AI and don't feel accountable for what the tooling is doing and that's one thing we totally need to work on as leaders, as managers, as colleagues, as users. I've seen some software I've been using definitely lacking quality over the last month. And I read in the press that those people are massively using AI and I'm like, uh, maybe I see something here and I'd like to know what kind of feedback I can give to get them back on quality.

08:43 >> Mhm. So, we're collapsing roles. >> Yep. So you mentioned the engineer of the tester. What about the architect? What about the product manager? What about the security tester, the compliance elements? The more people you have in an org, the more narrowed focus you have for people. When you're in a startup company, you are everybody. You know, the tester, the PM, the architects, and sometimes the user, the dog foodter, and the HR and the finance.

09:14 you're everybody, right? And the bigger the or grows and the more people you need in every single verticals and eventually you have a PM for the front end, a technical PM for the back end, you're going to have a a handful of software engineer for the front from the back for the cues and the more it grows and then it becomes like an army people working on it and with AI we try to shrink it back again to when the industry was growing to the company was growing and we face sometimes some misunderstanding of people who

09:44 have never been a PM even like a bad PM but when you have to be a PM because you're in a sub company you eventually do the job and when you the AI comes it really helps you when you've never been but you can be and you're asked to be that's something you have to learn I believe that we need some more trainings or help for all those people because AI really really empowers you but it empowers you to do things that you're not sure how to do and that's where AI barely helps you it always answer the question you're asking

10:15 but if you're asking the wrong questions and being a senior is asking the right questions the good question that helps grow that helps define that's being a senior senior is with experience and knowledge you actually ask the question that helps move forward to a solution and I believe we need to help people being better at asking the right questions to the AI if you really want to succeed >> it worries me that the software engineer who's been used to being asked to build this and given some sort of a spec, whether it's

10:49 a set of user stories, a screen mockup, or just a quick conversation, we're now asking that person to think of the right questions to ask. What is the implication of that? We've spoken about shift left and shift right for the last couple of decades. We're shifting further left. But have we and how do we empower and give the people who are being now asked to do the questioning to do that left shift to know how to do that work.

11:22 >> People dare to do things if they know they will be accountable but not blamed for what happens. And that's an organization. That's a human part of your organization. If you empower people but they don't dare you just not doing anything and that comes back to human to managers to organizations like how do you empower your people and give them not only the tooling but also the state of mind of stepping out of your comfort zone I'm a software engineer and as I said I have to also explore broader space what if I don't

11:55 dare what if I can't because even though I got all the tooling I have I fear it I fear the organization I work with and that's something which is this is a hard pause. I mean, humanly is the hard part of the software industry, right? >> So, this is culture. This is the engineering culture podcast. How do I nurture that culture? >> I'm not sure that the recipe.

12:15 I sorry, I know there's no recipe. However, I got some good examples where I believe it was the right move. The first thing is transparency and honesty. If you are honest with your team, most of the time people start being honest. I eventually became the CTO of a company which had a broken culture. People would lie, hide, steal, you know, everything was wrong and I was asked to fix the culture so to speak and very quickly I decided to, you know, lead by example.

12:50 So I started to have a Q&A every week. You can ask me any question. If I can answer, I answer. If I can't, you have to keep me accountable for answering the next time. And you ask me the question again. And if I don't answer, ask it again until you have the answer. No judgment about it. Of course, first Q&A, I only got easy questions. People wouldn't dare to challenge the head of the org.

13:14 And I started to come with answers such as, I don't know, I made a mistake. This is a good thing. Congratulations. But like, you don't name people. You only name solutions and product. You are accountable for what you're doing and you take the load whenever and coming back with the last time you asked me these questions and I found the answer and it is unpleasant but you need to hear it and we are all in this together and gathering the thing.

13:42 Eventually after a few sessions people started to ask me difficult questions and eventually we started to go with even more difficult questions by allowing for an anonymous board. So I ask my manager, what about you ask them if they have questions and if they have it on there, write it down and you ask it. Even better, you give it to another manager so I cannot even guess which team it comes from.

14:05 But I want the questions. I don't care who ask it. So please dare. And they started to organize this. So I could never guess from which part of the organization question was coming. But I got the questions. And eventually people started to vote for questions and it grew like this. And it becomes very interesting. Eventually, people would stack questions before the Q&A.

14:24 What I noticed in New York is when I write, people will lie to me and I would see it and I was like, "Okay, you can totally, you know, confront the person. You're lying to me. That's val." I said, "No, no, no. It's it's a cultural thing. How do I take it as a whole and shift almost the entire culture of running of you're hiding things by being transparent and explaining things and showing them that I can be accountable and move forward and fix things.

14:49 You know being exemplary is when you can totally break things as long as you fix them before and you learn a lesson and you don't repeat your mistakes. Being wrong and making mistakes is totally acceptable. What is not acceptable is repeating your own mistakes because you have not learned. So you have not grown. And this cultural thing of allowing people totally shift the essence of this organization which went much better by the time.

15:13 But you also have to be patient. When you write software, you see it right away under your hands, maybe a few hours later. With humans, you need to learn patience, but eventually you see results. >> Beautiful story and a great example of shifting culture. What if I'm not high enough up in the organization to do this? >> Okay, first, there's nothing as just a team lead.

15:42 You are a team lead. That's not just a team lead. And at every level you got something to contribute to and to help grow. I've seen it work in smaller teams. Eventually I was in a startup company, a startup. We were already 1,200 in the company. So in scale up you say and I found two teams who were already working like this at their level. They were already working on this like transparency and we are a team but also we also need to be accountable.

16:12 And I've seen it work. It was less efficient in terms of I would say impact because it was at the size of the one team and not the entire org but already very interesting and the good thing is managers also gathered together. I mean as a manager you have a team but you are part of another team within the team of managers and that's the difficulty of being a manager that most people don't get first is you're not only manager of a team you are no member of two teams with sometimes different goals opposite goals as a

16:44 manager you need to think outside of your own team to help the org at the size of every manager as a manager of a team you need to help the team grow And sometimes you have to make decisions that won't please either the rest of the managers or sometimes your team. And it cannot only be one way. You have to think outside of your own team to make it grow, make it better.

17:09 And that's maybe the most difficult part of being a manager. And at every level, you always have a way to influence right outside. If you start to do the right thing and you actually see benefits, you can talk about it to other managers and eventually they will follow you and do the same thing and eventually it will grow. Okay, this is, you know, wishful thinking maybe, but if you don't try, you know, it won't work, right?

17:34 >> Yeah. If you don't even try, you're guaranteed to fail. >> The way I see it is as a manager, I'm here to help people. So, I would always not maybe always phrase it this way, but I say help me help you. So if you provide the right information to your manager, your manager will be able to help you. If you hide too much, if you're too secret on some part of who you are, what you do, it becomes a little harder.

17:59 You're not only a machine coding or being an engine, you're also a human. And sometimes you got ups and downs. And if you provide some information to help your manager, your manager can also totally help you. And that works at every level. as a CTO, I was the comics and talking with the CEO, the CFO and if you give them some hints that they can help you because it's you're in a down moment or up moment or you're available to help them, you become a member of the team and it's not a day by day.

18:28 It's an average. It's going to be 50/50 in the end. But it doesn't have to be you always being 50% helped and being helpful every day, by the hour, by the week. It's overall, but it's kind of hard thing because you don't know when it ends. >> We're dealing with people. >> I like to help people. I like to see people grow. Most of the time stop relationship in helping people and that's all why I moved to the management and leadership roles because I believe that most of people are better than I am at what they do in

19:00 their job. So if I can help them being even better, it cascades to the next person and the overall organization grows. So the manager is amplifier >> and that's also why when you got toxic management it's devastating for an org. The impact is easier when you got people reporting to you but also as easier it's easier to to destroy everything or to grow better.

19:28 We've the handed a lot and given some good advice around that transparency and bringing that into culture and the culture of accountability without blame. What is the really important question that I haven't asked you? >> We've talked about management and the impact that come with the role. Is it possible to have the same kind of impact without being manager?

19:53 I've seen individuals being able to do this and I've been always very much impressed by their level of influence and I'm not sure I've deciphered how they were working. What I've seen is those people were technically very strong, very humble and the common characteristic of those people were like they were always always listening more than they were speaking.

20:20 So yes, it is possible, but I'm also leaning towards the conclusion that it's not everybody is able to do it. There's something natural in those people. And when you ask them, they're like, I don't know. I got it by chance. For some reason, people are listening to me. I'm like, oh, come on, dude. You know that's wrong. You work on your skills, but also you've met the right people.

20:42 You've everybody's been mentored. Nobody came naturally into this kind of position. That's the one thing is you can totally grow. It is possible and it can be a north star to have this kind of influence. And if you don't reach this kind of influence is maybe you're too humble to understand that you are actually at this level of influence and it is more probable that is your humility that forgives you to see it than the fact that you are not at this level of influence.

21:06 But yes, it is possible. It is a very tough career. Some of the best leaders I've met as I sees many many of them tried being manager for a while and came back and the easiest and best seniors staff principal engineers I've managed knew the role of managers and knew how to help me as a manager. So I would say give it a go. It doesn't have to last for your life.

21:31 I mean you can totally give it a try for a moment. I was a an IC at Microsoft, became a manager, was hired as an IC at Facebook, became a manager, eventually created my company, was an IC, then again a manager, and then was hired as director, CTO. I've been an IC for a bit more than a year. Going back and forth gives you all the understanding of the scope of what is it to create the product that answers a user's demand.

21:58 In the end, it's just about this. I believe >> that pendulum swing across. >> Yeah, >> you got a lot of really interesting and meaty stuff there. If people want to continue the conversation, where can they find me? >> They can find me on X Twitter as Mu French. My alias is Mu and on LinkedIn by my name Venturi. Uh please please come and discuss it all.

22:31 You agree, you disagree, you have another story, you have questions, you have things to raise questions from me because I'm very curious. Please keep this conversation going. I think it's a fascinating space and we've decided to be in this space because it is fascinating because the meeting of software and humans is an ongoing conversation that's going to be for life. >> You guys, thanks so much for taking the time. >> Thank you, Shane, for having me in your podcast. Very, very interesting conversation.